Picture sentiment classification method based on double-attention multi-layer feature fusion

Abstract

The invention discloses a picture sentiment classification method based on double-attention multi-layer feature fusion, and belongs to the technical field of image processing. The picture sentiment classification method based on double-attention multi-layer feature fusion comprises a multi-level feature extraction network, a double-attention mechanism and an attention feature fusion sentiment classification module. The method comprises the steps: firstly, extracting multi-level features of multiple channels of an image through the multi-level feature extraction network; secondly, endowing spatial attention weights to the multi-channel low-level features through a spatial attention mechanism, endowing channel attention weights to the multi-channel high-level features through a channel attention mechanism, and respectively strengthening feature representation of different levels; the method is reasonable in design, makes full use of the complementarity of different levels of features ofthe image, fully considers the spatial information of the features and the semantic difference of different channel features, and enhances the feature representation through an attention mechanism, thereby improving the effect of picture emotion classification.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a picture emotion classification method based on double-attention multi-layer feature fusion. Background technique [0002] Social networks play an important role in people's daily life. With the development of mobile terminal technology and the popularity of camera-capable devices, more and more social users create and share massive text, image and video content through social media every day. For these Content sentiment analysis is widely used in recommendation, advertising, public opinion monitoring and other fields. Image sentiment classification needs to solve the "emotional gap" between image visual features and emotional semantics, and sentiment analysis is extremely challenging due to the complexity and subjectivity of emotions. [0003] Existing studies have shown that image emotion is related to different levels of visual features of the image. Early image se...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a picture emotion classification method based on double-attention multi-layer feature fusion, to solve the problem that the existing deep learning-based picture emotion classification method proposed in the background technology fails to make full use of multi-level features and features The problem of lack of expressive ability

Images